Catalysts, membranes and electricity prices dominate cost discussions in water electrolysis. The product itself — the gas bubble — is usually treated as something that simply leaves. Three reviews say otherwise: at industrially relevant current densities, bubbles shape activation, ohmic and mass-transport losses simultaneously, and they are also where local drying, hot spots and mechanical damage begin. How much current an electrode can deliver depends heavily on how fast the bubbles leave.
This article draws on three reviews: Yuan et al. on bubble evolution, impact and management (Progress in Energy and Combustion Science, 2023); Zhang et al. on gas–liquid flow in the anode porous transport layer (International Journal of Hydrogen Energy, 2025); and Doan et al. on porous transport layer design (International Journal of Energy Research, 2021). All figures are taken from those papers with their test conditions noted; interpretations without a citation are our own.
1. A counter-intuitive starting point: too few bubbles, not too many
The usual intuition is that more bubbles means more blocked surface and worse transport. The literature says the opposite. Bubble evolution is driven by gas supersaturation near the catalytic surface; generating bubbles consumes dissolved oxygen and lowers the local O₂ concentration, which reduces mass-transport loss. Bubbles also induce micro-convection (the Marangoni effect) around the evolution site, which further helps transport.
Hence an explicit warning in the source review: “It is easy to mistakenly associate an increase in the number of bubbles with an increase in mass transport loss… the increase in mass transport loss is not caused by too many bubbles, but by too few.”
That sentence sets the direction of every design route that follows. When mass-transport loss rises, the real cause is that supersaturation is not being relieved by a high enough detachment frequency — too few nucleation sites, individual bubbles growing too large, detachment too slow. What you observe is the inverse: more bubbles visible at any instant (high detachment frequency) and smaller bubbles (small detachment diameter) correspond to a lower measured mass-transport overpotential.
Practical consequence: gas production rate is the wrong metric. Track nucleation site density, detachment diameter and detachment frequency. Every effective change to an electrode or a diffusion layer moves one of those three.
The three loss categories sit alongside this mechanism: bubbles masking catalyst surface cause activation loss; bubbles are non-conductive, so coverage of the electrode surface and accumulation inside the PTL raise ohmic resistance; and a gas barrier that keeps water from reaching the catalyst layer causes local drying — with consequences including large voltage differences between regions of the same cell at the same current density, simultaneous rises in mass-transport and ohmic loss, and hot spots where heat cannot be dissipated.
2. The full path of a bubble: two processes, two locations
The problem splits into two physical processes. Evolution happens at the catalytic surface: nucleation, growth, detachment. Transport happens across components and interfaces: bubbles move from the catalyst layer (CL) into the porous transport layer (PTL) and are then carried away by water in the flow channel. How long and how tortuous that path is depends on the porosity, wettability and structure of each component.
Two specific mechanisms explain why transport stalls, and they underpin almost every PTL improvement reported. First, in-plane drift: bubbles prefer to move from small pores to adjacent large pores, and because pore size and spatial distribution are random, a neighbouring large pore may sit in the in-plane direction, making the path tortuous and slowing transport. Second, under-rib trapping: some transport paths deliver gas beneath the flow-field rib, where out-of-plane transport is obstructed and gas accumulates.
The design criteria therefore become concrete: at a given current density, increase the number of nucleation sites, reduce the detachment diameter, and raise the detachment frequency — the last especially at high current density. A second review adds a mechanistic point: capillary force is the dominant mechanism of bubble transport inside PTL pores, and since PTL surface wettability is tunable, wettability and structural parameters can be changed together to lower mass-transport loss without sacrificing activation loss — provided the effect of current density is accounted for.
3. Four design routes, one objective
Grouped by which layer is being changed, the cases in the three reviews converge into four routes. They share the same objective — the three criteria above plus one path principle: give the gas a sufficient, direct, low-resistance way out.
3.1 Catalyst layer: get the three-phase interface right first
Ionomer content is the parameter that governs CL pore structure. Tomographic reconstruction combined with ionomer modelling shows that increasing ionomer content fills pore volume, lowers porosity and raises tortuosity; at high content, performance falls and mass-transport loss rises, indicating that low porosity and high tortuosity raise O₂ transport resistance. Too little ionomer also costs performance, through higher ionic resistance and the associated overpotential. It is a two-sided trade-off that needs an optimum.
Loading and thickness show a similar window. Raising Ir loading from 1 to 5 mg cm⁻² thickens the CL from about 5 μm to more than 20 μm. A thin, low-loading CL makes the PTL–CL interface less uniform, while a thick one impedes water and gas transport. The quantitative case: at 1.5 A cm⁻², a 2.2 mg cm⁻² sample shows a 23% mass-transport share of losses, rising to 69.9% at 0.45 mg cm⁻².
Superaerophobic electrodes are the most active direction. The criterion is a bubble contact angle typically above 150° with negligible hysteresis; microstructural roughness can amplify the contact angle of an intrinsically aerophobic surface into the superaerophobic range, and designing a surface microstructure that supports the Cassie state is a promising route. Reported systems are mostly array structures — nanosheets, nanoplates, nanodendrites, nanopillars, nanotubes, nanocones — with representative data in Table 1. The engineering difficulties are equally explicit: these synthesis routes are hard to apply directly on a membrane, Ti-foil samples are difficult to transfer intact by decal, and the route of growing the catalyst directly on the PTL runs into poor ionomer contact with all catalyst particles and the risk that added ionomer destroys the superaerophobic character.
Ordered three-dimensional structures pay off in three ways: higher effective surface area gives more nucleation sites; nucleation position can be controlled to reduce adhesion probability; and transport paths become unobstructed. A direct comparison comes from gas-injection tests: an unordered nickel foam electrode releases bubbles in intermittent bursts with a 3131 ms period, indicating accumulation, while a 3D-printed nickel electrode releases steadily at roughly 124 ms intervals. But the size window is narrow — the densest pores (around 200 nm) actually reduce mass and specific activity, and increasing pore size further to 1200 nm gives no improvement, while greater thickness degrades performance by lengthening the bubble path.
Catalyst-coated diffusion layers (CCLGDL) treat the interface itself as the design object: rather than bonding a PTL to a CCM, the CL is coated directly onto a thin titanium PTL and then laminated to the membrane, removing CL regions that the PTL cannot contact. The starting observation was visual: both hydrogen and oxygen bubbles form mainly at the interface between CL and adjacent PTL, and bubbles form only around good electrical conductors — showing that in-plane electronic conduction matters. The magnitude is worth remembering: a cell with 0.17 mg Ir cm⁻² performs comparably to 3 mg Ir cm⁻² CCM/Ti-felt, i.e. 16× Ir mass activity; on the Pt side, 0.086 mg cm⁻² matches 3 mg cm⁻², up to 31×.
3.2 Porous transport layer: replace random pores with designed paths
Patterned through-pores are the cleanest concept: pores run from the PTL–flow-field interface all the way to the CL–PTL interface, offering low resistance so that gas is guided along them. In-situ X-ray radiography confirms that although gas generation sites are distributed widely across the interface, most bubbles still reach the PTL–flow-field interface through the through-pores. The result: at 9.0 A cm⁻², gas saturation near the interface is about half that of a sintered titanium powder PTL, the out-of-plane saturation profile is uniform, and mass-transport loss falls by 76.7%. The cost is a slight increase in activation loss from reduced interface contact; pore pattern and diameter remain to be optimised.
Graded structures follow a blade-like logic: small pores near the CL maximise contact, pore size increases toward the flow-field interface so that gas exits along buoyancy with minimum resistance and shortest path, and the lower porosity at the base provides a good electronic path. Pore-network modelling plus single-cell validation point the same way: when porosity increases from the CL side toward the flow field (LP→HP), oxygen saturation drops markedly and liquid permeability rises; the reverse HP→LP gradient causes oxygen to accumulate near the interface. Multilayer PTLs belong to the same family — co-sintering a microporous layer with a support layer reduces surface roughness, improves bonding to a rough CL surface, more than doubles catalyst utilisation and cuts mass-transport loss by 45% at 2.0 A cm⁻².
Wettability modification is the cheapest lever with a meaningful effect. Moving from hydrophobic to hydrophilic reduces the contact angle and improves performance: a hydrophilic PTL (10°) shows about 200 mV lower voltage than a hydrophobic one (120°) at 2 A cm⁻². Superhydrophilic treatment (boiling in alkaline/oxidising solution) gives lower oxygen saturation under operando neutron imaging and more than 11% lower cell voltage above 5.0 A cm⁻², with the benefit growing at higher current. Further still, amphiphilic patterning builds hydrophobic and hydrophilic channels that separate gas from liquid transport: oxygen grows and detaches faster in the hydrophobic channels, and a cell with the patterned PTL reaches 3.2 A cm⁻² at 2 V against 1.4 A cm⁻² for the pristine PTL.
3.3 Flow field and cell architecture: remove an interface entirely
Conclusions on conventional flow fields remain contradictory: single-serpentine and parallel fields have both been reported as superior, while parallel-field bipolar plates have become more common in recent years — possibly because shorter channels let bubbles leave more easily. Other work uses structured mesh channels that break large bubbles apart and suppress slug and annular flow.
A more radical step is to remove the anode flow field. One design combines a sintered titanium porous layer with a titanium expanded mesh, with the PTL in direct contact with a flat plate; the mesh peaks and valleys act as the flow field and form a pore-size gradient. Eliminating the PTL–flow-field interface reduces the associated ohmic loss and, compared with a mesh-only cell, raises efficiency by 31% at 4.0 A cm⁻². Another integrates pin flow channels, PTL structure and catalyst layer into a single all-in-one bipolar electrode made by 3D printing plus sputter coating: ohmic resistance is about 100 mΩ cm², 1.62 V delivers 1000 mA cm⁻², and Pt mass activity reaches 4.48 A mg⁻¹.
3.4 Active management: treat bubbles as an operating variable
External fields and operating strategies also work, each within limits. For magnetic fields the orientation is decisive: a field perpendicular to a horizontal electrode causes bubbles to rotate, a field parallel to a vertical electrode accelerates detachment, but if the Lorentz force opposes buoyancy, the field becomes a force that hinders detachment. With proper arrangement, 0.5 T raises performance by 33%, and the gain is not evident at low water flow. Ultrasound improves detachment through cavitation: at 50 W and 60 Hz in alkaline media, efficiency rises 5–18% above 80 mA cm⁻², and total efficiency still improves after accounting for ultrasonic energy. Hyper-gravity from 25 G to 256 G lowers overpotential at 0.5 A cm⁻². Pulsed voltage exploits a pumping effect that fully detaches bubbles between two voltage states, lowering energy per unit hydrogen; pressure swing introduces periodic sudden pressure drops, where bubbles first grow rapidly to create a voltage peak and then detach, bringing voltage back down — the review judges this likely to be effective in PEMWE as well.
4. Putting the numbers side by side
A single paper's conclusion can mislead if read without its test conditions. Placing the representative results of the four routes on a common relative basis makes the magnitude gap obvious: mass-activity gains sit between two- and thirty-fold, while loss reductions mostly fall between 10% and 77%. These two classes of number are not interchangeable — one saves precious metal, the other saves electricity.
| Approach | Representative work | Key result | Condition |
|---|---|---|---|
| CCLGDL (thin Ti PTL, coated CL) | Zhang group | 0.17 mg Ir cm⁻² matches 3 mg cm⁻² CCM/Ti-felt, 16× mass activity; 0.086 mg Pt cm⁻², 31× | single cell |
| Patterned through-pore PTL | Bazylak group | Interface gas saturation about half that of sintered Ti powder PTL; mass transport −76.7% | 9.0 A cm⁻² |
| Multilayer PTL (MPL-PTL) | Schuler et al. | Catalyst utilisation >2×; mass transport −45% | 2.0 A cm⁻² |
| Superhydrophilic PTL (boiling treatment) | Zhao et al. | Lower oxygen saturation than untreated; cell voltage −11% or more, growing with current | >5.0 A cm⁻² |
| Amphiphilic patterned PTL | Lim et al. | 3.2 A cm⁻² at 2 V vs 1.4 A cm⁻² for pristine PTL | 2 V |
| Anode flow field removed (sintered layer + Ti mesh) | Stiber et al. | Efficiency +31% vs mesh-only cell | 4.0 A cm⁻² |
| All-in-one bipolar electrode | Yang et al. | Ohmic ~100 mΩ cm²; 1.62 V at 1000 mA cm⁻²; 4.48 A mg⁻¹ Pt | single cell |
| Superaerophobic ordered 3D CL (AEMWE) | Wan et al. | Mass-transport share of losses 32.3% → 9.6% | 2000 mA cm⁻² |
| Catalyst-coated TTLGDL (AEMWE) | Zhang group | Separation of kinetic and polarisation behaviour postponed from 0.3 to 1.6 A cm⁻² | — |
| Ionomer optimisation (pure-water AEMWE) | Wan et al. | 1.90 V at 1900 mA cm⁻², 2.2× a conventional IrO₂ CCM CL | pure water |
| Magnetic field, optimised arrangement | Kaya et al. | Performance +33%; gain not evident at low water flow | 0.5 T |
| Ultrasonic field | Li et al. | Efficiency +5–18% net of ultrasonic energy input | 50 W / 60 Hz, alkaline |
5. Extending to AEMWE: pure water adds one more constraint
Most PEMWE bubble-management strategies transfer to AEMWE — apart from component materials and liquid feed, the MEA and cell structure are nearly identical. Reported examples include building a vertical channel array facing the PTL with a superaerophobic surface, where mass-transport loss is almost zero at 500 mA cm⁻²; reducing PTL thickness from about 260 μm (Ti foam) to about 50 μm in a catalyst-coated thin foil, which postpones the deviation of the kinetic segment; 3D-printed ordered electrodes; and self-supported structures with catalyst grown directly on the PTL, avoiding the interface problems of a sprayed CL.
AEMWE, however, has a variable PEMWE does not: once operation switches from liquid electrolyte to pure water, the ionomer network becomes the only ionic pathway. The cost is concrete — without the additional ionic transport that KOH provides, ECSA utilisation drops and conventional self-supported electrodes degrade markedly. The effect of ionomer content is non-monotonic: at zero, pure-water performance falls sharply; 10% already improves it substantially; at 20% performance falls again, because the surface becomes over-covered, high-frequency resistance rises and electron transfer is impeded. The optimised electrode reaches 1900 mA cm⁻² at 1.90 V in pure water, 2.2× a conventional IrO₂ powder CL.
For teams working on pure-water AEMWE, the ionomer is both the ionic pathway and a valve on the bubble pathway. It sets ionic resistance, but it also changes morphology, contact angle and porosity — meaning it directly rewrites the conditions for nucleation and detachment, and may destroy an otherwise superaerophobic surface. Ionomer formulation therefore cannot be optimised against electrochemical metrics alone; it has to be evaluated together with bubble behaviour.
A second route is directly relevant to manufacturing: spraying a nickel-based microporous layer onto the PTL. A low-capillary-pressure MPL promotes water transport to the reaction sites and suppresses the reverse hydrogen–oxygen flow that masks active sites, while the denser coating also improves PTL–CL contact. Impedance data show gains from simultaneous reductions in charge-transfer resistance and mass-transport resistance — the same logic as multilayer PTLs on the PEMWE side.
6. Three sentences to take away
Ask which variable you changed. Any change to an electrode, diffusion layer or flow field should be able to state whether it moved nucleation site density, detachment diameter or detachment frequency, and whether the exit path got shorter or longer. If it cannot, it is still just “a structure we tried”.
What works is structuring, not homogenising. Graded pores, patterned through-pores, amphiphilic patterns, ordered 3D structures, self-supported electrodes — none of these works by making porosity more uniform; all replace random pores with designed paths. Which means process capability — coating density, pattern resolution, interfacial bonding — is itself the differentiator, and explains why performance data in this field scatter so widely.
In flexible operation, treat bubbles as a measurable variable. Gas saturation inside the PTL responds much more slowly than current, and repeated load cycling accumulates residual gas, while pulsed voltage and pressure swing have both been shown to accelerate detachment. That apparent contradiction suggests a practical angle: turn bubble state from a post-hoc performance penalty into something observable and adjustable in operation — potentially more differentiating for a wind- and solar-coupled electrolyser than another increment of voltage efficiency.
A note on the data: these results come from different laboratories, cell designs and test conditions, mostly small-area single cells. Direct comparison carries risk and does not imply proportional reproduction at megawatt scale. The benefit of bubble management depends strongly on flow-field design, water flow rate, operating temperature and current-density range, and must be re-validated during scale-up.
References
S. Yuan, et al. Bubble evolution and transport in PEM water electrolysis: Mechanism, impact, and management. Progress in Energy and Combustion Science, 2023, 96: 101075. DOI: 10.1016/j.pecs.2023.101075
X. Zhang, et al. A review of gas-liquid flow characteristics of anode porous transport layer in proton exchange membrane electrolysis cell. International Journal of Hydrogen Energy, 2025, 100: 1010–1029. DOI: 10.1016/j.ijhydene.2024.12.392
T. L. Doan, et al. A review of the porous transport layer in polymer electrolyte membrane water electrolysis. International Journal of Energy Research, 2021. DOI: 10.1002/er.6739 (open access)
About AEMHY
Changzhou AEMHY Hydrogen Energy Technology Co., LTD. develops anion exchange membrane (AEM) water electrolysis, with full-stack capability from non-precious-metal catalysts, ionomers, electrodes and membrane electrode assemblies through to electrolysers and hydrogen production systems. Products cover AEM catalysts and electrodes, anion exchange membranes and ionomers, QZC-series electrolysers, hydrogen generation systems and test benches.
To discuss ionomer formulation for AEM electrodes, diffusion layer selection, or bubble management under pure-water operation, contact us via www.aemhy.com. Contact: Feng Liu, +86 188 1316 9835.